Thread by @JamesBisbee

Hey #polisci Twitter! @jmlisin, @kmmunger and I analyzed you! We found a lot of interesting things. osf.io/dvkwt/ 1/18

Late in 2018, we identified the universe of political scientists working at PhD-granting institutions and determined which of them were on Twitter. We then scraped them in the spring of 2019. 2/18

#polisci Twitter is predominantly male, tenured, and liberal, consistent with the discipline in general. However, women and tenure-track scholars are more likely to be on Twitter than men and tenured scholars. 3/18

Furthermore, the vast majority of scholars follow a minority of other scholars, meaning most of us mostly follow non-academics (or at least non-#polisci Twitter academics). 4/18

There are about 7 main communities with 20 or more scholars who retweet each other, and the communities are surprisingly easily identified by the users' bios. 5/18

The share of under-represented groups in academia that we are able to measure are over-represented in these groups, relative to their share of the discipline writ large. 6/18

Also, these communities are not totally silo'ed, with strong evidence of retweeting reciprocity between them. 7/18

We don't find any systematic differences between men and women, ideology, and school characteristics in terms of node centrality. However, it does appear that tenure-track scholars are more central in the retweet network. 8/18

But there is strong evidence of homophily when we look at who mentions whom, with groups being significantly more likely to mention co-members than non-co-members. 9/18

When we run dyadic regressions (with dyad robust clustered errors @cdsamii!), the evidence of homophily declines, although gender is still a big exception. Men are far more likely to mention other men and less likely to mention women. 10/18

When we look at tweets about research, we find the homophily disappears, with the exception -- again -- of gender. Even controlling for followers, we find that male scholars who see research shared by both men and women are more likely to engage with the male tweets. 11/18

These behaviors don't influence the spread of research. The only covariates that predict who shares research, how it is retweeted / favorited, and how many potential followers see it, are tenure status and years online, with newer, junior scholars being more active. 12/18

Finally, we look at the content of tweets that mention other scholars using the peRspective package to measure toxicity. We find that tweets mentioning women, tenured scholars, and liberals are more likely to be toxic. 13/18

But these probabilities are quite low. Less than 3% of scholars are mentioned in tweets with greater than 50% chance of being toxic. Overall, #polisci Twitter is a relatively polite place, at least compared to the general public in October of 2016. 14/18

This draft has a lot of problems. We only look at scholars at PhD-granting schools so we don't include a number of very influential people like @JohnHolbein1 and @dandrezner , and influential accounts like @womenalsoknow and @monkeycageblog. 15/18

The draft is also too long, mainly due to throwing in a kitchen sink of results. It's also purely descriptive (although ideas for online experiments are percolating!) Finally, it is unclear whether methods for estimating ideology online can work on political scientists. 16/18

It's been exciting to work on though, and we're excited to see what #polisci thinks of this mirror we're holding up to it. 17/18

Lots of gratitude to @jrpjrpjrp, @JohnHolbein1, @maqartan, @andyguess, @p_barbera, and participants at the #APSA2018 Diversity and Inclusion Hackathon which prompted this research idea. 18/18

@jrpjrpjrp @JohnHolbein1 @maqartan @andyguess @p_barbera 19/18. These results are based on scholars employed at PhD-granting institutions **IN THE UNITED STATES**. I apologize to my non-US brethren for my sloppy language. These patterns describe a subset of a subset of #polisci Twitter. Thanks to @DrAdrianBlau for calling me out!